Convolutional Neural Network-Based Drone Detection and Classification Using Overlaid Frequency-Modulated...

Area of Science:

• Radar Systems Engineering
• Artificial Intelligence
• Aerospace Engineering

Background:

• Existing drone detection methods struggle with small or distant targets due to signal attenuation and faint micro-Doppler signatures (MDS).
• Limitations in current techniques necessitate advanced solutions for reliable drone identification.

Purpose of the Study:

• To propose a novel drone detection method using convolutional neural networks (CNNs) and frequency-modulated continuous-wave (FMCW) radar.
• To overcome the performance degradation issues associated with conventional micro-Doppler signature (MDS)-based methods.

Main Methods:

• Utilizing range-Doppler map images generated from FMCW radar.
• Overlaying multiple time-series range-Doppler images into a single image.
• Employing a convolutional neural network (CNN) for drone detection and classification.

Main Results:

• Demonstrated significant performance improvements in drone detection accuracy.
• Achieved higher accuracy compared to conventional drone detection methods.
• Validated the method using experimental data from three different drone sizes.

Conclusions:

• The proposed CNN-based method effectively enhances drone detection accuracy, particularly for challenging scenarios.
• This novel approach offers a robust alternative to traditional methods, addressing limitations with small and distant drones.
• The technique shows promise for improved surveillance and security applications.